Clinical and biochemical data
We recruited 495 HCC patients and randomly divided them into the training cohort (248 patients) and the validation cohort (247 patients). The mean postoperative follow-up time was 51.6 months (median, 46.0 months; range, 2.0 to 120.0 months). In the training and validation cohorts, the median age of patients was 49.33 and 50.96 years, respectively. The proportion of male patients was much higher than that of female patients, and there were 219 male cases (88.3%) in the training cohort and 213 male cases (86.2%) in the validation cohort, which may be caused by the higher proportion of male liver cancer patients in Asian countries. Clinical and biochemical data were further statistically compared between the training and validation cohorts. The results were shown in Table 1.
Table 1
Clinical and biochemical data of examined patients.
Parameter
|
Training cohort
|
Validation cohort
|
P value
|
(n = 248)
|
(n = 247)
|
Basic information
|
|
|
|
Age (years)
|
49.33 ± 11.35
|
50.96 ± 11.80
|
0.119
|
Gender: n (%)
|
female
|
29 (11.7)
|
34 (13.8)
|
0.489
|
male
|
219 (88.3)
|
213 (86.2)
|
|
Family history: n (%)
|
no
|
216 (87.0)
|
219 (88.7)
|
0.435
|
yes
|
32 (13.0)
|
28 (11.3)
|
|
Drinking: n (%)
|
no
|
140 (56.5)
|
133 (53.8)
|
0.560
|
yes
|
108 (43.5)
|
114 (46.2)
|
|
Smoking: n (%)
|
no
|
148 (59.7)
|
152 (61.5)
|
0.577
|
yes
|
100 (40.3)
|
95 (38.5)
|
|
HBsAg: n (%)
|
negative
|
41 (16.5)
|
34 (13.8)
|
0.391
|
positive
|
207 (83.5)
|
213 (86.2)
|
|
Lab check data
|
|
|
|
WBC (× 109/L)
|
6.04 ± 2.01
|
6.39 ± 2.20
|
0.061
|
NEUT (× 109/L)
|
3.65 ± 1.71
|
3.92 ± 1.73
|
0.081
|
LYMPH (× 109/L)
|
1.64 ± 0.57
|
1.73 ± 0.65
|
0.105
|
Platelets (× 109/L)
|
173.96 ± 75.18
|
181.23 ± 79.27
|
0.096
|
Albumin (g/L)
|
39.07 ± 4.59
|
39.65 ± 4.66
|
0.220
|
Globulin (g/L)
|
31.01 ± 5.80
|
30.35 ± 6.18
|
0.215
|
TBIL (µmol/L)
|
15.91 ± 14.33
|
16.45 ± 16.07
|
0.747
|
DBIL (µmol/L)
|
6.33 ± 12.38
|
6.92 ± 13.17
|
0.689
|
ALT (U/L)
|
45.12 ± 42.93
|
51.08 ± 46.33
|
0.783
|
AST (U/L)
|
49.98 ± 48.83
|
51.91 ± 57.49
|
0.697
|
ALP (U/L)
|
95.69 ± 65.26
|
92.29 ± 42.60
|
0.493
|
GGT (U/L): median, range
|
67.62, 10.7-335.1
|
72.19, 10.0-351.76
|
0.854
|
AGLR level: median, range
|
90.63, 19.43-441.72
|
88.83, 16.07-462.16
|
0.521
|
AFP (ng/ml): median, range
|
246.7, 0.20-32800
|
220.7, 0.60-25410
|
0.363
|
Pathological features
|
|
|
|
Cirrhosis: n (%)
|
no
|
24 (10.0)
|
13 (5.3)
|
0.062
|
yes
|
224 (90.0)
|
234 (94.7)
|
|
Tumor size (cm)
|
7.81 ± 4.68
|
7.12 ± 4.11
|
0.085
|
Tumor number: n (%)
|
single
|
190 (76.6)
|
188 (76.1)
|
0.896
|
multiple
|
58 (23.4)
|
59 (23.9)
|
|
TNM stage: n (%)
|
I-II
|
136 (54.8)
|
124 (50.2)
|
0.302
|
III-IV
|
112 (45.2)
|
123 (49.8)
|
|
MVI: n (%)
|
no
|
201 (81.0)
|
188 (76.1)
|
0.181
|
yes
|
47 (19.0)
|
59 (23.9)
|
|
Recurrence: n (%)
|
no
|
158 (63.7)
|
148 (59.9)
|
0.385
|
yes
|
90 (36.3)
|
99 (40.1)
|
|
N, number of patients; HBsAg, hepatitis B surface antigen; WBC, white blood cell; LYMPH, lymphocyte count; TBIL, total bilirubin; DBIL, direct bilirubin; ALT, alanine aminotransferase; AST, aspartate aminotransferase; ALP, alkaline phosphatase; GGT, gamma-glutamyl transpeptidase; AGLR, ALP plus GGT to LYMPH; AFP, alpha-fetoprotein; TNM, tumor-node-metastasis; MVI, microvascular invasion. |
The relationship between preoperative AGLR level and clinical pathologic characteristics in patients with HCC
Using the receiver operator characteristics (ROC) analysis, the optimal predictive cut-off value of AGLR was 90, with the sensitivity of 75.1%, the specificity of 64.8% and the area under the curve (AUC) was 0.735 (95% CI: 0.679–0.786), according to the postoperative survival of HCC patients in the training cohort. Based on this cut-off value, our patients could be divided into two groups by dichotomy: AGLR ≤ 90 and AGLR > 90 groups. Given that serum AFP level is a prognostic factor of liver cancer either, thus we performed a comparison analysis between AFP and AGLR. Interestingly, it was revealed that the AUCs of AGLR were higher than that of AFP in both training cohort and validation cohort (Fig. 1A, S1A). Meanwhile, C-index of AGLR suggested that both AGLR (C-index = 0.637, 95%CI, 0.597–0.684) and AFP (C-index = 0.624, 95%CI, 0.585–0.671) had predictive value in the training cohort, and more importantly, AGLR had a higher accuracy than AFP; and the value of AGLR (C-index = 0.654, 95%CI, 0.613–0.707) and AFP (C-index = 0.577, 95%CI, 0.532–0.633) were both verified in the validation cohort. The relationships between preoperative AGLR level and clinicopathologic characteristics were investigated and results were shown in Table 2. In the training cohort (248 patients), high preoperative AGLR level was positively correlated with serum AFP level (> 20 ng/ml) (p < 0.001), tumor size > 5 cm (p < 0.001), multiple tumors (χ2 = 86.367, p = 0.035), TNM stage III-IV (p < 0.001), presence of MVI (p < 0.001). And in the validation cohort (247 patients), high preoperative AGLR level was positively correlated with serum AFP level (> 20 ng/ml) (p < 0.001), tumor size > 5 cm (p < 0.001), TNM stage III-IV (p < 0.001), presence of MVI (p < 0.001). However, there were no obvious correlations between AGLR > 90 and age, gender, drinking, HBsAg, liver cirrhosis and recurrence (all p > 0.05). Moreover, higher AGLR level was found in tumor size > 5 cm, TNM stage III-IV and MVI patients (p < 0.05, Fig. 1B, S1B). These results suggested that elevated serum AGLR level may be related to poor progression and microvascular invasion of HCC.
Table 2
Correlation between clinical pathologic characteristics and AGLR level in HCC patients.
Variables
|
AGLR level
|
Training cohort (n = 248)
|
Validation cohort (n = 247)
|
≤ 90 n (%)
|
> 90 n (%)
|
P value
|
≤ 90 n (%)
|
> 90 n (%)
|
P value
|
Age (years)
|
≤ 60
|
92 (46.2)
|
107 (53.8)
|
0.347
|
81 (41.3)
|
115 (58.7)
|
0.815
|
> 60
|
19 (38.8)
|
30 (61.2)
|
|
22 (43.1)
|
29 (56.9)
|
|
Gender
|
Female
|
18 (60.0)
|
12 (40.0)
|
0.073
|
18 (52.9)
|
16 (47.1)
|
0.152
|
Male
|
93 (42.7)
|
125 (57.3)
|
|
85 (39.9)
|
128 (60.1)
|
|
Drinking
|
No
|
62 (44.6)
|
77 (55.4)
|
0.958
|
58 (43.3)
|
76 (56.7)
|
0.583
|
Yes
|
49 (45.0)
|
60 (55.0)
|
|
45 (39.8)
|
68 (60.2)
|
|
HBsAg
|
Negative
|
14 (38.9)
|
22 (61.1)
|
0.444
|
17 (43.6)
|
22 (56.4)
|
0.794
|
Positive
|
97 (45.8)
|
115 (54.2)
|
|
86 (41.3)
|
122 (58.7)
|
|
AFP (ng/ml)
|
≤ 20
|
54 (62.8)
|
32 (37.2)
|
< 0.001
|
43 (58.6)
|
30 (41.1)
|
< 0.001
|
> 20
|
57 (35.2)
|
105 (64.8)
|
|
60 (34.5)
|
114 (65.5)
|
|
Liver cirrhosis
|
No
|
8 (57.1)
|
6 (42.9)
|
0.337
|
10 (43.5)
|
13 (56.5)
|
0.856
|
Yes
|
103 (44.0)
|
131 (56.0)
|
|
93 (41.5)
|
131 (58.5)
|
|
Tumor size (cm)
|
≤ 5
|
66 (65.3)
|
35 (34.7)
|
< 0.001
|
61 (56.5)
|
47 (43.5)
|
< 0.001
|
> 5
|
45 (30.6)
|
102 (69.4)
|
|
42 (30.2)
|
97 (69.8)
|
|
Tumor number
|
Single
|
90 (48.6)
|
95 (51.4)
|
0.035
|
84 (43.5)
|
109 (56.5)
|
0.272
|
Multiple
|
21 (33.3)
|
42 (66.7)
|
|
19 (35.2)
|
35 (64.8)
|
|
TNM stage
|
I- II
|
82 (68.9)
|
37 (31.1)
|
< 0.001
|
81 (57.4)
|
60 (42.6)
|
< 0.001
|
III- IV
|
29 (22.5)
|
100 (77.5)
|
|
22 (20.8)
|
84 (79.2)
|
|
Microvascular invasion
|
No
|
95 (52.2)
|
87 (47.8)
|
< 0.001
|
96 (46.2)
|
112 (53.8)
|
0.001
|
Yes
|
16 (24.2)
|
50 (75.8)
|
|
7 (17.9)
|
32 (82.1)
|
|
Recurrence
|
No
|
70 (47.0)
|
79 (53.0)
|
0.388
|
71 (45.2)
|
86 (54.8)
|
0.138
|
Yes
|
41 (41.4)
|
58 (58.6)
|
|
32 (35.6)
|
58 (64.4)
|
|
AGLR, alkaline phosphatase plus gamma-glutamyl transpeptidase to lymphocyte ratio; HBsAg, hepatitis B surface antigen, AFP, alpha-fetoprotein, TNM, tumor-node-metastasis. |
Survival analysis based on different preoperative AGLR levels
In the training cohort, the average survival time for DFS patients with AGLR ≤ 90 was 77.42 months (95% CI, 67.70-87.13), and for DFS patients with AGLR > 90, the average survival time was 39.52 months (95% CI, 31.90-47.15) (p < 0.001, Fig. 2A). Among OS patients, the average survival time of patients with AGLR ≤ 90 was 83.60 months (95% CI, 75.18–92.03) and the 1-, 3- and 5-year survival rates were 86.3%, 71.6% and 63.1%, respectively; while for AGLR > 90 patients, they had an average OS of 47.39 months (95% CI, 40.26–54.53) and the 1-, 3- and 5-year survival rates were 77.6%, 44.8% and 27.4%, respectively (p < 0.001, Fig. 2B).
In the validation cohort, the average survival time for DFS patients with AGLR ≤ 90 was 75.58 months (95% CI, 66.12–85.03), and for patients with AGLR > 90, the average survival time was 49.28 months (95% CI, 41.42–57.14) (p < 0.001; Fig. S2A). In OS patients, for HCC patients with AGLR ≤ 90, the average survival time was 83.66 months (95% CI, 75.65–91.66) and the 1-, 3- and 5-year survival rates were 88.7%, 73.0% and 60.8%, respectively; and for patients whose AGLR > 90, they had a mean OS of 59.30 months (95% CI, 52.10–66.50) and the 1-, 3- and 5-year survival rates were 73.5%, 46.9% and 36.1%, respectively (p < 0.001, Fig. S2B). Therefore, the results clearly suggested that high AGLR level may predict poor prognosis for HCC patients.
Prognostic factors of survival for patients with HCC
The Cox univariate and multivariate regression analyses were applied to evaluate the prognostic value of AGLR and other factors. In the training cohort, it was found that AGLR > 90 (HR = 1.79, 95% CI, 1.21–2.69, p < 0.001), tumor size (HR = 1.91, 95% CI, 1.27–2.61, p < 0.001), TNM stage (HR = 1.52, 95% CI, 1.03–2.31, p = 0.025), MVI (HR = 1.61, 95% CI, 1.23–2.39, p = 0.007) and recurrence (HR = 2.01, 95% CI, 1.47–2.83, p < 0.001) were five crucial independent predictors of OS for HCC patients (Table 3), and similar result was found in the validation cohort either (Table S1).
Table 3
Univariate and multivariate analysis of overall survival (training cohort, n = 248).
Clinical character
|
Univariate analysis
|
Multivariate analysis
|
HR (95% CI)
|
p value
|
HR (95% CI)
|
p value
|
AGLR level (> 90 vs ≤ 90)
|
2.66 (2.01–3.88)
|
< 0.001
|
1.79 (1.21–2.69)
|
< 0.001
|
Age, years (> 60 vs ≤ 60)
|
1.24 (0.81–1.83)
|
0.308
|
|
|
Gender (male vs female)
|
1.23 (0.72–1.91)
|
0.441
|
|
|
Drinking (yes vs no)
|
1.03 (0.74–1.41)
|
0.862
|
|
|
HBsAg (positive vs negative)
|
1.29 (0.78–2.06)
|
0.303
|
|
|
AFP, ng/ml (> 20 vs ≤ 20)
|
1.71 (1.19–2.47)
|
0.003
|
1.13 (0.77–1.67)
|
0.514
|
Liver cirrhosis(yes vs no)
|
1.03 (0.51–2.09)
|
0.930
|
|
|
Tumor size, cm (> 5 vs ≤ 5)
|
2.86 (1.97–3.91)
|
< 0.001
|
1.91 (1.27–2.61)
|
< 0.001
|
Tumor number (multiple vs single)
|
1.60 (1.13–2.26)
|
0.006
|
1.12 (0.81–1.53)
|
0.460
|
TNM stage (III–IV vs I–II)
|
1.96 1.39–2.77)
|
< 0.001
|
1.52 (1.03–2.31)
|
0.025
|
MVI (yes vs no)
|
2.57 (1.93–3.74)
|
< 0.001
|
1.61 (1.23–2.39)
|
0.007
|
Recurrence (yes vs no)
|
2.70 (1.69–3.59)
|
< 0.001
|
2.01 (1.47–2.83)
|
< 0.001
|
CI, confidence interval; HR, hazard ratio; AGLR, alkaline phosphatase plus gamma-glutamyl transpeptidase to lymphocyte ratio; HBsAg, hepatitis B surface antigen; AFP, alpha-fetoprotein; TNM, tumor-node-metastasis; MVI, microvascular invasion. |
Then, each of the above five independent predictors were assigned, such as AGLR ≤ 90 was assigned 0 point and AGLR > 90 was assigned 1 point, and other four predictors were assigned in the same manner. Thus, all HCC patients would be divided into six groups of different scores, ranging from 0 to 5 points, based on their accumulated total scores. As the result, a new prognostic scoring model consisted of multiple variables was constructed. However, some comparisons between two of these new groups had no statistical different. For instance, in the training cohort, for DFS patients with a score of 2 vs. 3 (p = 0.173) (Fig. 3A) and for OS patients with a score of 2 vs. 3 (p = 0.126), score 3 vs. 4 (p = 0.062) and score 4 vs. 5 (p = 0.079) (Fig. 3B). Similar result was also found in the validation cohort (Fig. S3A-B). In view of these circumstances and in order to obtain better application value of this new model, we further divided these groups according to the scores: 0–1 points (low-risk group), 2–3 points (medium-risk group) and 4–5 points (high-risk group). Surprisingly, the survival analyses revealed that in both training cohort (Fig. 3C-D) and validation cohort (Fig. S3C-D), HCC patients’ postoperative survival time had significant differences between the low-, medium- and high-risk groups, which was an obviously decreasing trend, suggesting that this novel scoring model may have potential application value in predicting postoperative prognosis for liver cancer patients.